Insurance doesn't look like an AI industry. But sit with any policy long enough and you find the same thing underneath: information. Applications, risk models, claims files, photographs, recorded calls, policy documents, pricing tables, regulatory requirements. Mountains of it. Which makes insurance one of the better-suited industries for AI, whether or not it looks the part. The money says the industry agrees. Gartner forecast that worldwide insurance IT spending would reach $227.7 billion in 2025, up 7.9% from the year before. KPMG found that 73% of insurance CEOs ranked AI as a top investment priority.
Adoption is moving just as fast. Goldman Sachs Asset Management's 2026 Global Insurance Survey found the share of insurers actively using AI rose from 29% in 2024 to 48% in 2025 to 62% in 2026 — with another 34% considering it. None of this is trend-chasing. Insurers see a chance to rebuild an industry that runs on information, prediction, paperwork and human judgment. So where is the money actually going — and what changes for the person holding the policy?
Why Insurance Fits AI So Well
Insurers spend their days estimating risk, pricing uncertainty, spotting patterns, reading documents, chasing anomalies, answering complicated questions and pushing claims through. That list is close to a description of what modern AI does well.
The industry also runs on an enormous amount of manual work. An underwriter might pull information from applications, internal databases, email threads and supporting documents before getting anywhere near an actual decision. Accenture estimates that roughly 40% of an average underwriter's time goes to administrative and other non-core tasks.
The same friction shows up in claims, service and policy administration. Documents live in different systems. Employees re-key information the company already has somewhere else. Customers dig through a sixty-page policy document, give up, and call an agent to ask something basic — like whether their deductible resets in January. Multiply that across millions of policies and claims, and small improvements stop being small.
McKinsey puts the opportunity at roughly $50 billion to $70 billion in revenue impact for the industry from generative AI, with the biggest openings in marketing and sales, customer operations and software engineering. In a survey of more than 50 insurers, over half expected productivity gains of 10% to 20%. Hence the shift from side-project pilots to line items in the budget.
Where the Money Is Going
1. Underwriting: Getting to the Decision Faster
Most of the underwriting happens before anyone decides anything. Information has to be gathered, sorted and read across applications, prior claims, outside databases and whatever documents came attached. AI can pull the relevant data out, summarize a submission, notice what's missing and flag risks worth a closer look. Executives expect this to move quickly. Accenture found senior underwriting leaders anticipating AI adoption climbing from roughly 14% to 70% within three years.
That's not underwriters being replaced. The shape most insurers are aiming for is AI around human judgment, software absorbs repetitive information work, experienced people spend their time on complex cases, exceptions and calls that need context. Faster decisions, same accountability.
2. Claims and Fraud: Seeing Patterns People Can't
A single claim can generate a startling pile of material: forms, photos, repair estimates, invoices, medical records, message threads, sometimes sensor or location data. AI can read across several of those at once — prioritizing claims, catching inconsistencies, estimating damage, flagging patterns that look off. Deloitte estimates that AI applied across the property-and-casualty claims lifecycle could cut fraudulent losses by $80 billion to $160 billion by 2032.
Some insurers are already past the pilot stage. Allianz's Project Nemo uses seven specialized AI agents to handle straightforward food-spoilage claims after natural catastrophes, running coverage checks and fraud screening among other steps, and taking processing time from days down to hours. A human still makes the final payment decision. That last detail matters more than it sounds. Speed on simple claims is a real win, but the decisions with consequences still need review, accountability and a way to escalate when something doesn't fit the pattern.
3. Customer Service: Turning Policy Documents Into Conversations
This is the piece most people will actually feel. Insurance is hard largely because the information you need is buried — in exclusions, deductibles, network rules, coverage limits and vocabulary nobody uses anywhere else in life. The last wave of insurance technology moved all of that online. Useful, but the burden stayed with you: you still had to know where to look and what to look for.
Generative AI changes the interaction itself. Instead of searching for information, you ask about it: Is this doctor on the network? What is this bill actually charging me for? Have I hit my deductible? What's the real difference between these two plans? What should I check before I file a claim?
Search becomes conversation. The underlying information doesn't go anywhere — AI becomes the layer that finds it, organizes it and explains it in terms of what you asked. For a product as tangled as insurance, that's a bigger change than putting one more PDF on a website.
4. Distribution and Personalization: Help Before You Buy, Not Just After
AI is also moving earlier in the journey, to the part where someone is still figuring out what they need. That means reading what a customer is actually asking for, surfacing products that fit, explaining how plans differ, building comparisons on the spot, and helping agents find answers faster. It matters because most insurance shoppers don't know which questions they're supposed to be asking in the first place.
The old path looked like this: Browse → Compare → Call → Buy. The emerging one looks more like this: Ask → Understand → Compare → Act. For anyone who grew up expecting to just ask, that's a far more natural way in than clicking through menus, brochures and product pages.
The Real Reason Insurers Are Investing: AI Changes the Economics
Three incentives sit behind all of this.
Lower operating cost: Still the clearest one. Goldman Sachs' 2026 survey found that 83% of insurers using or considering AI named reducing operational costs as an application — the most frequently selected use case. Across underwriting, claims, servicing and back-office work, AI cuts down the hours spent retrieving information, reading routine documents and re-entering data the company already has. At insurance scale, minutes compound.
Better growth: This is the newer part. In Accenture's 2026 survey of 263 senior insurance executives, 81% of organizations reported at least a 5% improvement in gross written premiums tied to their data and AI initiatives, through pricing, personalization and cross-selling. AI has stopped being purely a back-office cost story. It's now an acquisition, retention and distribution story too.
Competitive pressure: Then there's the cost of sitting still. BCG reports that AI spending as a share of revenue among P&C insurers is expected to roughly triple from 2025 to 2026, with early movers positioned for 3% to 5% premium growth. Once a few carriers make quoting, servicing or simple claims genuinely painless, that becomes the baseline customers expect from everyone else. Spending on AI is partly an offense, partly insurance against falling behind.
What This Looks Like From the Customer Side
Before anyone buys anything, finding out what a policy covers means searching a site, opening a PDF, and reading past language written for regulators rather than for you. Comparing two plans means lining up deductibles and exclusions by hand and hoping you didn't skip the clause that matters. In an AI-assisted version, both become questions you ask. What does this cover if I'm driving a rental car? The work of assembling the answer moves off the customer.
Support changes shape too. Today you wait for an agent whether your question was complicated or not. Put a first-line assistant ahead of the queue and simple questions stop consuming the wait — people who need a human still get one, just sooner.
Language is the piece that gets undersold. Most of the U.S. insurance experience is still built around English documents, and people navigating it in a second language have spent years being handed those documents and told to figure it out. Multilingual interaction isn't a convenience for them. It's the difference between understanding what you bought and guessing.
Further along, the shifts get less visible and larger: claims move from sequential manual review to automated triage with human oversight, so a simple file doesn't sit behind a complex one. Personalization stops meaning which segment are you in and starts meaning something closer to your circumstances.
But More AI Doesn't Automatically Mean Better Insurance
Insurance is high-stakes, and AI now touches what products you see, how your risk gets evaluated and how your claim gets handled. That raises real questions about fairness, privacy, accuracy and accountability. Historical data carries historical patterns, and models that are poorly designed or left unmonitored can carry them forward. Insurance files hold some of the most sensitive personal, financial and health information a person has. And generative AI can produce answers that sound authoritative and are simply wrong.
U.S. regulators are watching closely. The National Association of Insurance Commissioners' Model Bulletin on AI is direct about it: decisions made or supported by AI remain subject to existing insurance law. It puts weight on governance, risk management and processes built to produce accurate, fair outcomes for consumers. Insurers are building those controls too. EY found that 79% of surveyed insurers were pursuing monitoring and auditing of AI data alongside clear privacy and security policies aimed at accuracy and fairness. The systems that win here probably won't be the ones automating the most decisions. They'll be the ones that combine intelligence, reliable data, guardrails and human judgment.
From Digital Insurance to Intelligent Insurance
The past decade put insurance online. Websites replaced paper forms, apps replaced some phone calls, portals gave people access to their policies, quotes and claims. What's coming next is a different kind of change. Insurance becomes intelligent: systems stop simply displaying information and start helping people understand it — retrieving what's relevant, comparing options, working across languages, pointing to the next step.
There's a long way to go. BCG found that only 38% of P&C insurers are currently generating AI value at scale across core workflows, a reminder that large budgets and working implementations are not the same thing. But the direction is hard to miss. The prize isn't running the same insurance processes slightly faster. It's rethinking how people deal with insurance at all.
At Tigerless AI, that's a large part of what we're working on with Lara: an AI assistant built to help people ask questions, make sense of insurance information and find their way through a system that was never designed to be easy to understand. The last generation of insurance technology put insurance online.
At Tigerless AI, the part of this we find most interesting isn't the underwriting and claims work the budgets are mostly chasing. It's the consumer end — someone trying to understand a policy, a bill, or a decision that's already been made about them. That's where the gap between what the industry has built and what people actually experience is widest, and it's the part that gets the least attention in reports like the ones quoted above.
One thing has become clear watching this space: capability stopped being the hard problem a while ago. The boundary is the hard problem. An assistant that explains how a deductible works is useful. An assistant that sounds equally confident explaining why a claim was denied is a liability. The two answers read identically. The stakes don't. Getting clear about which questions belong with a licensed person is a design decision, not a gap to be engineered away in a later version.
Which is where the rest of this piece keeps landing. The insurers who get real value out of this won't be the ones automating the most decisions. They'll be the ones who worked out which decisions shouldn't be. The last generation of insurance technology put insurance online. It didn't make insurance any easier to understand. That's the part still worth building.
